Assignment 2 Master Solution. Part (a) /ESD.77 Spring 2004 Multidisciplinary System Design Optimization. (a1) Decomposition

Size: px
Start display at page:

Download "Assignment 2 Master Solution. Part (a) /ESD.77 Spring 2004 Multidisciplinary System Design Optimization. (a1) Decomposition"

Transcription

1 16.888/ESD.77 Spring 004 Multidisciplinary System Design Optimization Assignment Master Solution Part (a) (a1) Decomposition The following system of equations must be solved: xx x + = 0 (1) 1 3 x + 3x 9 = 0 () 5 x x x x + 10 = 0 (3) x 3x + 7 = 0 (4) 5 xx xx + x 9 = 0 (5) 5 4 1) All-at-once First, we solve for a solution, numerically, all at once (MATLAB version 5): Script (Aa1.m) % Aa1: solve A part (a1) clear all close all eqn1='x1*x-*x3+'; eqn='x+3*x5-9'; eqn3='x1-x4*x5-x3+10'; eqn4='9*x5-3*x+7'; eqn5='x*x5-x*x4+x-9'; flops(0) tic [x1,x,x3,x4,x5]=solve(eqn1,eqn,eqn3,eqn4,eqn5) toc flops The following solution is returned: T T x = x1 x x3 x4 x5 = = [ ] [ ] The computational effort is: CPU time = 0.19 [sec] FLOPS = 3795 T

2 ) Decomposition The second method decomposes the system of equations into smaller systems and solves them sequentially. First we find the occurrence matrix of the system: X1 X X3 X4 X5 E1 x x x E x x E3 x x x x E4 x x E5 x x x Based on this we reorder the occurrence matrix to mae it a lower left triangular blocdiagonal matrix: X X5 X4 X3 X1 E x x E4 x x E5 x x x E3 x x x x E1 x x x The solution strategy now is: 1: First we solve for x and x 5 from Eq. and Eq.4 : Next we solve for x 4 from Eq.5 3: Then we solve for x 3 and x 1 from Eq.3 and Eq.1 The following script solves this system: clear all close all flops(0) tic %first step x5=inv([1 3; -3 9])*[9-7]'; x=x5(1); x5=x5(); %second step x4=(9-x-x*x5)/(-x); %third step x13=inv([x -; 1-1])*[- -10+x4*x5]'; x1=x13(1); x3=x13(); x=[x1 x x3 x4 x5]' toc flops The solution is obtained as:

3 x = The computational effort is: CPU time = 0.01 [sec] FLOPS = 134 Conclusion: The solution by decomposition is less computationally expensive by a factor of ~19 in CPU time and a factor of ~8 in floating point operations. In this special case no iterations are required. This effect can be magnified for large systems of coupled, nonlinear equations. The solution by decomposition, however, requires more upfront wor and understanding of the structure of the system of equations. (a) Gradient-Based Optimization Rosenbroc s banana function is given as: (, ) = ( 1 ) + 100( ) f x x x x x This test function often causes gradient based optimizers problems, due to it s numerical conditioning. This is the main point of this problem. It is called the banana function because of the way the curvature bends around the origin. It is notorious in optimization examples because of the slow convergence which most methods exhibit when trying to solve this problem. This function has a unique minimum at (1,1): f(1,1)=0.

4 The gradient vector can be calculated analytically 1 as: T f f 3 f = = 400x1 + ( 1 00x) x1 00( x x1 ) x1 x There are no constraints, g, h in this case and the KKT optimality conditions reduce to ( ) * * x : f x = 0 We can verify that the point (1,1) indeed meets this condition by substitution. The Hessian matrix, H, is obtained as: f f x x x x ( x ) x H = = f f 400x1 00 x x1 x At (1,1) the Hessian matrix is T f f x1 x1 x H = = f f x x x 1 (1,1) The eigenvalues of H at (1,1) are: λ = λ = Both eigenvalues are positive, therefore the point (1,1) is at least a local minimum. Note that the Rosenbroc banana function is not convex, so global optimality cannot be proven, because H is not semi-positive definite (SPD) for all setting of x. The point (1,1) is the nown global minimum of this function, since a sum of squares is minimized when all individual terms of the sum go to zero. The problem for numerically searching for this solution is that the two eigenvalues are three orders of magnitude apart. This is said to be a numerically ill-conditioned situation. Most gradient based optimization algorithms, start form an initial point, x o, and try to reach a point where the KKT conditions are true in an iterative fashion: 1 Rather than approximating the gradient vector by finite differencing.

5 x = x + α S 1 where α is the step size and S is the step direction vector. What distinguishes most algorithms is how α and S are computed throughout the search space. Depending on the method used to compute step size and step direction the algorithm will converge faster or slower for different types of ill-conditioned or non-convex functions. Or it might fail to converge altogether. Some of you might have experienced this problem initially. (i) Implement steepest gradient search and determine min(f) using at least two initial guesses: We select the following far-apart starting points: [-5,-3] and [4,] In steepest descent the step direction is chosen as the (local) direction that causes a maximal decrease of J (or f) at x: Choose 0 o x, set x= x Repeat until converged: 1 S = J x ( ) choose α to minimize J ( x+ αs ) update current point: x=x+ αs This algorithm doesn t use any information from previous iterations. The step size a is chosen with a 1D-search. Here a bisection algorithm is implemented, where α [0, αmax ] and α / mid = αmax. The size of this interval is adjusted, i.e. increased or decreased, until f ( x + α ) ( mid S < f x + αmaxs ) is true. Next a nd order equation (parabola) is fitted to f ( α ) with the three points 0, αmid, α max. The minimum of this parabola is chosen as the optimal step size α at the -th iteration. The line search for α is one of the most challenging aspects of writing a robust gradient search algorithm. The implementation code for steepest gradient search is enclosed below. Note: It is more robust to normalize the step direction to be a unit vector: 1 S = J x / J x 1 and have the step size be purely determined by α. ( ) ( ) (ii) Implement conjugate gradient search and determine min(f) using the same initial guesses as in (i) Conjugate gradient search is similar to steepest descent, with the exception that information from previous steps is used to compute the next step (direction), S. Here conjugate gradient search was implemented as follows:

6 S f ( x ) ( x ) f ( x ) 1 ( x ) = + β S f 1 β = f Note, that this means that the new step direction is the steepest gradient direction at iteration, modified by a contribution from the step direction, S, at the previous iteration -1. Thus, some memory is used without having to store the entire past matrix (history) of gradients. The contribution of the past step is scaled with a parameter β, which is larger if the norm of the gradient vector is larger for the current step,, and smaller if the norm of the gradient vector at is smaller than at -1. As we approach the (local or global) minimum we expect the norm of f to decrease, such that β<1 will generally be true in the vicinity of such a stationary point. (iii) Verify computational results against the analytical solution, and compare the performance of the two numerical methods. The analytical optimum is (1,1) as discussed above. The numerical results, as well as the expense for obtaining them, is compared in the following table and figure combinations for the two starting points. General conclusions about the behavior of both algorithms are also provided. The residual is simply the -norm (Euclidian norm) of the gradient vector: f f residual = f ( x ) = + x1 x x x The termination criterion was set as: residual < Conclusions Both methods are able to approximate the analytical optimum (1,1), albeit at different computational cost. For the starting point (-5,-3) the solution (1,1) is reached from below, for the starting point (4,) it is reached from above. The conjugate gradient method converges faster than steepest descent by a factor of 16-3 in terms of the number of function evaluations (iterations), CPU time and floating point operations (FLOPS). The exact savings factor depends on the start point. This is a very significant difference, which results mainly from the fact that conjugate gradient avoids zig-zaging in the design space the way steepest gradient does. This is achieved by smoothing the search direction with the conjugate directions from previous steps. Overall lesson: even small algorithmic changes can have a large (beneficial or detrimental) impact on the behavior of search methods, particularly for numerically ill-conditioned objective functions and constraints.

7 Results for Starting Point: (-5,-3) Steepest Gradient Search Starting Point: -5-3 Solution: Iterations: 1558 CPU time: 1.79 FLOPS: Residual: Conjugate Gradient Search Starting Point: -5-3 Solution: Iterations: 79 CPU time: 0.11 FLOPS: Residual: Figure: x o =[-5, -3]: Comparison of search path with steepest gradient (white dashed) and conjugate gradient algorithm (yellow dotted line).

8 Results for Starting Point: (4,) Steepest Gradient Search Starting Point: 4 Solution: Iterations: 54 CPU time: 6.49 FLOPS: Residual: Conjugate Gradient Search Starting Point: 4 Solution: Iterations: 169 CPU time: 0.1 FLOPS: Residual: Figure: x o =[4 ]: Comparison of search path with steepest gradient (white dashed) and conjugate gradient algorithm (yellow dotted line).

9 Gradient Optimization Code: function aa % Find minimum of Rosenbroc's "banana" function using two % different gradient search methods: steepest gradient % search and conjugate gradient search % dwo, March 004 clear all close all warning off % plot function dx = 1/5; [x1,x] = meshgrid(-5:dx:5); f=(1-x1).^+100.*(x-x1.^).^; surf(x1,x,f) view ([ 8]) axis([min(min(x1)) max(max(x1)) min(min(x)) max(max(x))... min(min(f)) max(max(f))]) xlabel('x_1'), ylabel('x_'), zlabel('f'), title('rosenbroc Banana Function') drawnow % start optimization xi=[-5-3]; x=xi; f=banana(x(1),x()); epsilon=1e-3; % steepest gradient search =1; T1=cputime; F1=flops; residual=1; alpha=1; while residual>epsilon % compute gradient gradf=[400*x(,1)^3+*(1-00*x(,))*x(,1)- 00*(x(,)- x(,1)^)]; S=-1/norm(gradf,)*gradf; %normalized step direction residual=norm(gradf,); % line search for optimal step size alpha - bisection method % use last alpha as initial guess alpha=[0 alpha *alpha]; foundalpha=0; while foundalpha==0; for inda=1:length(alpha) xsearch=x(,:)+alpha(inda)*s; fsearch(inda)=banana(xsearch(1), xsearch()); end if fsearch(3)<=fsearch() & fsearch()<=fsearch(1) %disp('expand search interval over alpha') alpha(1)=0; alpha(3)=*alpha(3); alpha()=alpha(3)/; elseif fsearch()>=fsearch(1) %disp('shrin search interval over alpha')

10 fit alpha(1)=0; alpha(3)=alpha(); alpha()=alpha(3)/; elseif fsearch()<=fsearch(1) & fsearch(3)>=fsearch() % interpolate for optimal alpha [Palpha,Salpha] = polyfit(alpha,fsearch,); % nd order alphafullsearch=linspace(alpha(1),alpha(3),1e3); ffullsearch=polyval(palpha,alphafullsearch,salpha); [tmp,inda]=min(fsearch); [fmin,inda]=min(ffullsearch); alphaopt=alphafullsearch(inda); %disp('found optimal alpha') foundalpha=1; end end %finished line search for alpha alpha=alphaopt; %alphasearch=[0:0.1:1.0]; %for inda=1:length(alphasearch) % xsearch=x(,:)+alphasearch(inda)*s; % fsearch(inda)=banana(xsearch(1), xsearch()); %end % polynomial fit for f(alpha) % [Palpha,Salpha] = polyfit(alphasearch,fsearch,); % nd order fit % alphafullsearch=[0:0.001:1.0]; % for inda=1:length(alphafullsearch) % ffullsearch(inda)=polyval(palpha,alphafullsearch(inda),salpha); % end % [fmin,inda]=min(ffullsearch); % alpha=alphafullsearch(inda) % update current solution x=x(,:); xplusone=x+alpha*s; x=[x; xplusone]; =+1; f()=banana(xplusone(1), xplusone()); end %while residual>epsilon 1=; T=cputime; F=flops; disp('steepest Gradient Search') disp(['starting Point: ' numstr(xi)]) disp(['solution: ' numstr(x(,:))]) disp(['iterations: ' numstr()]) disp(['cpu time: ' numstr(t-t1)]) disp(['flops: ' numstr(f-f1)]) disp(['residual: ' numstr(residual)]) % plot steepest gradient search path hold on plot3(x(:,1),x(:,),f,'w*') plot3(x(:,1),x(:,),f,'w--','linewidth', ) drawnow % conjugate gradient search xi=[-5-3]; x=xi; f=banana(x(1),x());

11 =1; T1=cputime; F1=flops; residual=1; alpha=1; gradf=[]; residual=1; S=[]; while residual>epsilon % compute gradient gradf(,:)=[400*x(,1)^3+*(1-00*x(,))*x(,1)- 00*(x(,)- x(,1)^)]; residual=norm(gradf(,:),); if ==1 S=-1/norm(gradf,)*gradf; %normalized step direction %normalized step direction else beta()=norm(gradf(,:),)^/norm(gradf(-1,:),)^; S(,:)=-(1/norm(gradf(,:),))*gradf(,:)+beta()*S(-1,:); S(,:)=S(,:)/norm(S(,:),); end fit % line search for optimal step size alpha - bisection method % use last alpha as initial guess alpha=[0 alpha *alpha]; foundalpha=0; while foundalpha==0; for inda=1:length(alpha) xsearch=x(,:)+alpha(inda)*s(,:); fsearch(inda)=banana(xsearch(1), xsearch()); end if fsearch(3)<=fsearch() & fsearch()<=fsearch(1) %disp('expand search interval over alpha') alpha(1)=0; alpha(3)=*alpha(3); alpha()=alpha(3)/; elseif fsearch()>=fsearch(1) %disp('shrin search interval over alpha') alpha(1)=0; alpha(3)=alpha(); alpha()=alpha(3)/; elseif fsearch()<=fsearch(1) & fsearch(3)>=fsearch() % interpolate for optimal alpha [Palpha,Salpha] = polyfit(alpha,fsearch,); % nd order alphafullsearch=linspace(alpha(1),alpha(3),1e3); ffullsearch=polyval(palpha,alphafullsearch,salpha); [tmp,inda]=min(fsearch); [fmin,inda]=min(ffullsearch); alphaopt=alphafullsearch(inda); %disp('found optimal alpha') foundalpha=1; end end %finished line search for alpha alpha=alphaopt; x=x(,:); xplusone=x+alpha*s(,:); x=[x; xplusone]; =+1; f()=banana(xplusone(1), xplusone()); end %while residual>epsilon 1=; T=cputime; F=flops; disp('conjugate Gradient Search') disp(['starting Point: ' numstr(xi)]) disp(['solution: ' numstr(x(,:))]) disp(['iterations: ' numstr()]) disp(['cpu time: ' numstr(t-t1)]) disp(['flops: ' numstr(f-f1)])

12 disp(['residual: ' numstr(residual)]) % plot steepest gradient search path hold on plot3(x(:,1),x(:,),f,'y*') plot3(x(:,1),x(:,),f,'y:','linewidth', )

Today. Golden section, discussion of error Newton s method. Newton s method, steepest descent, conjugate gradient

Today. Golden section, discussion of error Newton s method. Newton s method, steepest descent, conjugate gradient Optimization Last time Root finding: definition, motivation Algorithms: Bisection, false position, secant, Newton-Raphson Convergence & tradeoffs Example applications of Newton s method Root finding in

More information

Modern Methods of Data Analysis - WS 07/08

Modern Methods of Data Analysis - WS 07/08 Modern Methods of Data Analysis Lecture XV (04.02.08) Contents: Function Minimization (see E. Lohrmann & V. Blobel) Optimization Problem Set of n independent variables Sometimes in addition some constraints

More information

Introduction to optimization methods and line search

Introduction to optimization methods and line search Introduction to optimization methods and line search Jussi Hakanen Post-doctoral researcher jussi.hakanen@jyu.fi How to find optimal solutions? Trial and error widely used in practice, not efficient and

More information

Iterative Algorithms I: Elementary Iterative Methods and the Conjugate Gradient Algorithms

Iterative Algorithms I: Elementary Iterative Methods and the Conjugate Gradient Algorithms Iterative Algorithms I: Elementary Iterative Methods and the Conjugate Gradient Algorithms By:- Nitin Kamra Indian Institute of Technology, Delhi Advisor:- Prof. Ulrich Reude 1. Introduction to Linear

More information

(Sparse) Linear Solvers

(Sparse) Linear Solvers (Sparse) Linear Solvers Ax = B Why? Many geometry processing applications boil down to: solve one or more linear systems Parameterization Editing Reconstruction Fairing Morphing 2 Don t you just invert

More information

2. Linear Regression and Gradient Descent

2. Linear Regression and Gradient Descent Pattern Recognition And Machine Learning - EPFL - Fall 2015 Emtiyaz Khan, Timur Bagautdinov, Carlos Becker, Ilija Bogunovic & Ksenia Konyushkova 2. Linear Regression and Gradient Descent 2.1 Goals The

More information

arxiv: v1 [cs.cv] 2 May 2016

arxiv: v1 [cs.cv] 2 May 2016 16-811 Math Fundamentals for Robotics Comparison of Optimization Methods in Optical Flow Estimation Final Report, Fall 2015 arxiv:1605.00572v1 [cs.cv] 2 May 2016 Contents Noranart Vesdapunt Master of Computer

More information

Gradient, Newton and conjugate direction methods for unconstrained nonlinear optimization

Gradient, Newton and conjugate direction methods for unconstrained nonlinear optimization Gradient, Newton and conjugate direction methods for unconstrained nonlinear optimization Consider the gradient method (steepest descent), with exact unidimensional search, the Newton method and the conjugate

More information

Computational Optimization. Constrained Optimization Algorithms

Computational Optimization. Constrained Optimization Algorithms Computational Optimization Constrained Optimization Algorithms Same basic algorithms Repeat Determine descent direction Determine step size Take a step Until Optimal But now must consider feasibility,

More information

Introduction to Optimization Problems and Methods

Introduction to Optimization Problems and Methods Introduction to Optimization Problems and Methods wjch@umich.edu December 10, 2009 Outline 1 Linear Optimization Problem Simplex Method 2 3 Cutting Plane Method 4 Discrete Dynamic Programming Problem Simplex

More information

A Comparative Study of Frequency-domain Finite Element Updating Approaches Using Different Optimization Procedures

A Comparative Study of Frequency-domain Finite Element Updating Approaches Using Different Optimization Procedures A Comparative Study of Frequency-domain Finite Element Updating Approaches Using Different Optimization Procedures Xinjun DONG 1, Yang WANG 1* 1 School of Civil and Environmental Engineering, Georgia Institute

More information

Theoretical Concepts of Machine Learning

Theoretical Concepts of Machine Learning Theoretical Concepts of Machine Learning Part 2 Institute of Bioinformatics Johannes Kepler University, Linz, Austria Outline 1 Introduction 2 Generalization Error 3 Maximum Likelihood 4 Noise Models 5

More information

Lecture 8. Divided Differences,Least-Squares Approximations. Ceng375 Numerical Computations at December 9, 2010

Lecture 8. Divided Differences,Least-Squares Approximations. Ceng375 Numerical Computations at December 9, 2010 Lecture 8, Ceng375 Numerical Computations at December 9, 2010 Computer Engineering Department Çankaya University 8.1 Contents 1 2 3 8.2 : These provide a more efficient way to construct an interpolating

More information

Lecture 6 - Multivariate numerical optimization

Lecture 6 - Multivariate numerical optimization Lecture 6 - Multivariate numerical optimization Björn Andersson (w/ Jianxin Wei) Department of Statistics, Uppsala University February 13, 2014 1 / 36 Table of Contents 1 Plotting functions of two variables

More information

Lecture 12: Feasible direction methods

Lecture 12: Feasible direction methods Lecture 12 Lecture 12: Feasible direction methods Kin Cheong Sou December 2, 2013 TMA947 Lecture 12 Lecture 12: Feasible direction methods 1 / 1 Feasible-direction methods, I Intro Consider the problem

More information

Multivariate Numerical Optimization

Multivariate Numerical Optimization Jianxin Wei March 1, 2013 Outline 1 Graphics for Function of Two Variables 2 Nelder-Mead Simplex Method 3 Steepest Descent Method 4 Newton s Method 5 Quasi-Newton s Method 6 Built-in R Function 7 Linear

More information

Convex Optimization CMU-10725

Convex Optimization CMU-10725 Convex Optimization CMU-10725 Conjugate Direction Methods Barnabás Póczos & Ryan Tibshirani Conjugate Direction Methods 2 Books to Read David G. Luenberger, Yinyu Ye: Linear and Nonlinear Programming Nesterov:

More information

Programming 1. Script files. help cd Example:

Programming 1. Script files. help cd Example: Programming Until now we worked with Matlab interactively, executing simple statements line by line, often reentering the same sequences of commands. Alternatively, we can store the Matlab input commands

More information

Approximation Methods in Optimization

Approximation Methods in Optimization Approximation Methods in Optimization The basic idea is that if you have a function that is noisy and possibly expensive to evaluate, then that function can be sampled at a few points and a fit of it created.

More information

CPSC 340: Machine Learning and Data Mining. Robust Regression Fall 2015

CPSC 340: Machine Learning and Data Mining. Robust Regression Fall 2015 CPSC 340: Machine Learning and Data Mining Robust Regression Fall 2015 Admin Can you see Assignment 1 grades on UBC connect? Auditors, don t worry about it. You should already be working on Assignment

More information

Report of Linear Solver Implementation on GPU

Report of Linear Solver Implementation on GPU Report of Linear Solver Implementation on GPU XIANG LI Abstract As the development of technology and the linear equation solver is used in many aspects such as smart grid, aviation and chemical engineering,

More information

Numerical Solution to the Brachistochrone Problem

Numerical Solution to the Brachistochrone Problem Numerical Solution to the Brachistochrone Problem Aditya Mittal * Stanford University Abstract This paper presents the numerical solution to the brachistochrone problem. The brachistochrone problem is

More information

Contents. I Basics 1. Copyright by SIAM. Unauthorized reproduction of this article is prohibited.

Contents. I Basics 1. Copyright by SIAM. Unauthorized reproduction of this article is prohibited. page v Preface xiii I Basics 1 1 Optimization Models 3 1.1 Introduction... 3 1.2 Optimization: An Informal Introduction... 4 1.3 Linear Equations... 7 1.4 Linear Optimization... 10 Exercises... 12 1.5

More information

Programs. Introduction

Programs. Introduction 16 Interior Point I: Linear Programs Lab Objective: For decades after its invention, the Simplex algorithm was the only competitive method for linear programming. The past 30 years, however, have seen

More information

Numerical Optimization: Introduction and gradient-based methods

Numerical Optimization: Introduction and gradient-based methods Numerical Optimization: Introduction and gradient-based methods Master 2 Recherche LRI Apprentissage Statistique et Optimisation Anne Auger Inria Saclay-Ile-de-France November 2011 http://tao.lri.fr/tiki-index.php?page=courses

More information

Classical Gradient Methods

Classical Gradient Methods Classical Gradient Methods Note simultaneous course at AMSI (math) summer school: Nonlin. Optimization Methods (see http://wwwmaths.anu.edu.au/events/amsiss05/) Recommended textbook (Springer Verlag, 1999):

More information

Matlab (Matrix laboratory) is an interactive software system for numerical computations and graphics.

Matlab (Matrix laboratory) is an interactive software system for numerical computations and graphics. Matlab (Matrix laboratory) is an interactive software system for numerical computations and graphics. Starting MATLAB - On a PC, double click the MATLAB icon - On a LINUX/UNIX machine, enter the command:

More information

(Sparse) Linear Solvers

(Sparse) Linear Solvers (Sparse) Linear Solvers Ax = B Why? Many geometry processing applications boil down to: solve one or more linear systems Parameterization Editing Reconstruction Fairing Morphing 1 Don t you just invert

More information

Good luck! First name Legi number Computer slabhg Points

Good luck! First name Legi number Computer slabhg Points Surname First name Legi number Computer slabhg... Note 1 2 4 5 Points Fill in the cover sheet. (Computer: write the number of the PC as printed on the table). Leave your Legi on the table. Switch off your

More information

Introduction. Optimization

Introduction. Optimization Introduction to Optimization Amy Langville SAMSI Undergraduate Workshop N.C. State University SAMSI 6/1/05 GOAL: minimize f(x 1, x 2, x 3, x 4, x 5 ) = x 2 1.5x 2x 3 + x 4 /x 5 PRIZE: $1 million # of independent

More information

Energy Minimization -Non-Derivative Methods -First Derivative Methods. Background Image Courtesy: 3dciencia.com visual life sciences

Energy Minimization -Non-Derivative Methods -First Derivative Methods. Background Image Courtesy: 3dciencia.com visual life sciences Energy Minimization -Non-Derivative Methods -First Derivative Methods Background Image Courtesy: 3dciencia.com visual life sciences Introduction Contents Criteria to start minimization Energy Minimization

More information

Iterative Methods for Solving Linear Problems

Iterative Methods for Solving Linear Problems Iterative Methods for Solving Linear Problems When problems become too large (too many data points, too many model parameters), SVD and related approaches become impractical. Iterative Methods for Solving

More information

Introduction to Optimization

Introduction to Optimization Introduction to Optimization Second Order Optimization Methods Marc Toussaint U Stuttgart Planned Outline Gradient-based optimization (1st order methods) plain grad., steepest descent, conjugate grad.,

More information

Contents. I The Basic Framework for Stationary Problems 1

Contents. I The Basic Framework for Stationary Problems 1 page v Preface xiii I The Basic Framework for Stationary Problems 1 1 Some model PDEs 3 1.1 Laplace s equation; elliptic BVPs... 3 1.1.1 Physical experiments modeled by Laplace s equation... 5 1.2 Other

More information

An Adaptive Stencil Linear Deviation Method for Wave Equations

An Adaptive Stencil Linear Deviation Method for Wave Equations 211 An Adaptive Stencil Linear Deviation Method for Wave Equations Kelly Hasler Faculty Sponsor: Robert H. Hoar, Department of Mathematics ABSTRACT Wave Equations are partial differential equations (PDEs)

More information

Kernels and representation

Kernels and representation Kernels and representation Corso di AA, anno 2017/18, Padova Fabio Aiolli 20 Dicembre 2017 Fabio Aiolli Kernels and representation 20 Dicembre 2017 1 / 19 (Hierarchical) Representation Learning Hierarchical

More information

Concept of Curve Fitting Difference with Interpolation

Concept of Curve Fitting Difference with Interpolation Curve Fitting Content Concept of Curve Fitting Difference with Interpolation Estimation of Linear Parameters by Least Squares Curve Fitting by Polynomial Least Squares Estimation of Non-linear Parameters

More information

The Alternating Direction Method of Multipliers

The Alternating Direction Method of Multipliers The Alternating Direction Method of Multipliers With Adaptive Step Size Selection Peter Sutor, Jr. Project Advisor: Professor Tom Goldstein October 8, 2015 1 / 30 Introduction Presentation Outline 1 Convex

More information

Generic descent algorithm Generalization to multiple dimensions Problems of descent methods, possible improvements Fixes Local minima

Generic descent algorithm Generalization to multiple dimensions Problems of descent methods, possible improvements Fixes Local minima 1 Lecture 10: descent methods Generic descent algorithm Generalization to multiple dimensions Problems of descent methods, possible improvements Fixes Local minima Gradient descent (reminder) Minimum of

More information

Multidimensional scaling

Multidimensional scaling Multidimensional scaling Lecture 5 Alexander & Michael Bronstein tosca.cs.technion.ac.il/book Numerical geometry of non-rigid shapes Stanford University, Winter 2009 Cinderella 2.0 2 If it doesn t fit,

More information

6. Linear Discriminant Functions

6. Linear Discriminant Functions 6. Linear Discriminant Functions Linear Discriminant Functions Assumption: we know the proper forms for the discriminant functions, and use the samples to estimate the values of parameters of the classifier

More information

Model parametrization strategies for Newton-based acoustic full waveform

Model parametrization strategies for Newton-based acoustic full waveform Model parametrization strategies for Newton-based acoustic full waveform inversion Amsalu Y. Anagaw, University of Alberta, Edmonton, Canada, aanagaw@ualberta.ca Summary This paper studies the effects

More information

Lecture 18: March 23

Lecture 18: March 23 0-725/36-725: Convex Optimization Spring 205 Lecturer: Ryan Tibshirani Lecture 8: March 23 Scribes: James Duyck Note: LaTeX template courtesy of UC Berkeley EECS dept. Disclaimer: These notes have not

More information

CHAPTER 3 AN OVERVIEW OF DESIGN OF EXPERIMENTS AND RESPONSE SURFACE METHODOLOGY

CHAPTER 3 AN OVERVIEW OF DESIGN OF EXPERIMENTS AND RESPONSE SURFACE METHODOLOGY 23 CHAPTER 3 AN OVERVIEW OF DESIGN OF EXPERIMENTS AND RESPONSE SURFACE METHODOLOGY 3.1 DESIGN OF EXPERIMENTS Design of experiments is a systematic approach for investigation of a system or process. A series

More information

Constrained and Unconstrained Optimization

Constrained and Unconstrained Optimization Constrained and Unconstrained Optimization Carlos Hurtado Department of Economics University of Illinois at Urbana-Champaign hrtdmrt2@illinois.edu Oct 10th, 2017 C. Hurtado (UIUC - Economics) Numerical

More information

Programming, numerics and optimization

Programming, numerics and optimization Programming, numerics and optimization Lecture C-4: Constrained optimization Łukasz Jankowski ljank@ippt.pan.pl Institute of Fundamental Technological Research Room 4.32, Phone +22.8261281 ext. 428 June

More information

Efficient Iterative Semi-supervised Classification on Manifold

Efficient Iterative Semi-supervised Classification on Manifold . Efficient Iterative Semi-supervised Classification on Manifold... M. Farajtabar, H. R. Rabiee, A. Shaban, A. Soltani-Farani Sharif University of Technology, Tehran, Iran. Presented by Pooria Joulani

More information

Multi Layer Perceptron trained by Quasi Newton learning rule

Multi Layer Perceptron trained by Quasi Newton learning rule Multi Layer Perceptron trained by Quasi Newton learning rule Feed-forward neural networks provide a general framework for representing nonlinear functional mappings between a set of input variables and

More information

f xx + f yy = F (x, y)

f xx + f yy = F (x, y) Application of the 2D finite element method to Laplace (Poisson) equation; f xx + f yy = F (x, y) M. R. Hadizadeh Computer Club, Department of Physics and Astronomy, Ohio University 4 Nov. 2013 Domain

More information

Sequential Coordinate-wise Algorithm for Non-negative Least Squares Problem

Sequential Coordinate-wise Algorithm for Non-negative Least Squares Problem CENTER FOR MACHINE PERCEPTION CZECH TECHNICAL UNIVERSITY Sequential Coordinate-wise Algorithm for Non-negative Least Squares Problem Woring document of the EU project COSPAL IST-004176 Vojtěch Franc, Miro

More information

David G. Luenberger Yinyu Ye. Linear and Nonlinear. Programming. Fourth Edition. ö Springer

David G. Luenberger Yinyu Ye. Linear and Nonlinear. Programming. Fourth Edition. ö Springer David G. Luenberger Yinyu Ye Linear and Nonlinear Programming Fourth Edition ö Springer Contents 1 Introduction 1 1.1 Optimization 1 1.2 Types of Problems 2 1.3 Size of Problems 5 1.4 Iterative Algorithms

More information

DS Machine Learning and Data Mining I. Alina Oprea Associate Professor, CCIS Northeastern University

DS Machine Learning and Data Mining I. Alina Oprea Associate Professor, CCIS Northeastern University DS 4400 Machine Learning and Data Mining I Alina Oprea Associate Professor, CCIS Northeastern University September 20 2018 Review Solution for multiple linear regression can be computed in closed form

More information

Machine Learning for Signal Processing Lecture 4: Optimization

Machine Learning for Signal Processing Lecture 4: Optimization Machine Learning for Signal Processing Lecture 4: Optimization 13 Sep 2015 Instructor: Bhiksha Raj (slides largely by Najim Dehak, JHU) 11-755/18-797 1 Index 1. The problem of optimization 2. Direct optimization

More information

Search direction improvement for gradient-based optimization problems

Search direction improvement for gradient-based optimization problems Computer Aided Optimum Design in Engineering IX 3 Search direction improvement for gradient-based optimization problems S Ganguly & W L Neu Aerospace and Ocean Engineering, Virginia Tech, USA Abstract

More information

Lecture VIII. Global Approximation Methods: I

Lecture VIII. Global Approximation Methods: I Lecture VIII Global Approximation Methods: I Gianluca Violante New York University Quantitative Macroeconomics G. Violante, Global Methods p. 1 /29 Global function approximation Global methods: function

More information

CPSC 340: Machine Learning and Data Mining. Principal Component Analysis Fall 2016

CPSC 340: Machine Learning and Data Mining. Principal Component Analysis Fall 2016 CPSC 340: Machine Learning and Data Mining Principal Component Analysis Fall 2016 A2/Midterm: Admin Grades/solutions will be posted after class. Assignment 4: Posted, due November 14. Extra office hours:

More information

Computer Project 3. AA Computational Fluid Dyanmics University of Washington. Mishaal Aleem March 17, 2015

Computer Project 3. AA Computational Fluid Dyanmics University of Washington. Mishaal Aleem March 17, 2015 Computer Project 3 AA 543 - Computational Fluid Dyanmics University of Washington Mishaal Aleem March 17, 2015 Contents Introduction........................................... 1 3.1 Grid Generator.......................................

More information

Image Deconvolution.

Image Deconvolution. Image Deconvolution. Mathematics of Imaging. HW3 Jihwan Kim Abstract This homework is to implement image deconvolution methods, especially focused on a ExpectationMaximization(EM) algorithm. Most of this

More information

Characterizing Improving Directions Unconstrained Optimization

Characterizing Improving Directions Unconstrained Optimization Final Review IE417 In the Beginning... In the beginning, Weierstrass's theorem said that a continuous function achieves a minimum on a compact set. Using this, we showed that for a convex set S and y not

More information

Note Set 4: Finite Mixture Models and the EM Algorithm

Note Set 4: Finite Mixture Models and the EM Algorithm Note Set 4: Finite Mixture Models and the EM Algorithm Padhraic Smyth, Department of Computer Science University of California, Irvine Finite Mixture Models A finite mixture model with K components, for

More information

Convex Optimization. Erick Delage, and Ashutosh Saxena. October 20, (a) (b) (c)

Convex Optimization. Erick Delage, and Ashutosh Saxena. October 20, (a) (b) (c) Convex Optimization (for CS229) Erick Delage, and Ashutosh Saxena October 20, 2006 1 Convex Sets Definition: A set G R n is convex if every pair of point (x, y) G, the segment beteen x and y is in A. More

More information

Robot Mapping. TORO Gradient Descent for SLAM. Cyrill Stachniss

Robot Mapping. TORO Gradient Descent for SLAM. Cyrill Stachniss Robot Mapping TORO Gradient Descent for SLAM Cyrill Stachniss 1 Stochastic Gradient Descent Minimize the error individually for each constraint (decomposition of the problem into sub-problems) Solve one

More information

Numerical Modelling in Fortran: day 6. Paul Tackley, 2017

Numerical Modelling in Fortran: day 6. Paul Tackley, 2017 Numerical Modelling in Fortran: day 6 Paul Tackley, 2017 Today s Goals 1. Learn about pointers, generic procedures and operators 2. Learn about iterative solvers for boundary value problems, including

More information

Nonparametric Risk Attribution for Factor Models of Portfolios. October 3, 2017 Kellie Ottoboni

Nonparametric Risk Attribution for Factor Models of Portfolios. October 3, 2017 Kellie Ottoboni Nonparametric Risk Attribution for Factor Models of Portfolios October 3, 2017 Kellie Ottoboni Outline The problem Page 3 Additive model of returns Page 7 Euler s formula for risk decomposition Page 11

More information

Optimal Design of a Parallel Beam System with Elastic Supports to Minimize Flexural Response to Harmonic Loading

Optimal Design of a Parallel Beam System with Elastic Supports to Minimize Flexural Response to Harmonic Loading 11 th World Congress on Structural and Multidisciplinary Optimisation 07 th -12 th, June 2015, Sydney Australia Optimal Design of a Parallel Beam System with Elastic Supports to Minimize Flexural Response

More information

Deep Learning with Tensorflow AlexNet

Deep Learning with Tensorflow   AlexNet Machine Learning and Computer Vision Group Deep Learning with Tensorflow http://cvml.ist.ac.at/courses/dlwt_w17/ AlexNet Krizhevsky, Alex, Ilya Sutskever, and Geoffrey E. Hinton, "Imagenet classification

More information

Enclosures of Roundoff Errors using SDP

Enclosures of Roundoff Errors using SDP Enclosures of Roundoff Errors using SDP Victor Magron, CNRS Jointly Certified Upper Bounds with G. Constantinides and A. Donaldson Metalibm workshop: Elementary functions, digital filters and beyond 12-13

More information

ADAPTIVE APPROACH IN NONLINEAR CURVE DESIGN PROBLEM. Simo Virtanen Rakenteiden Mekaniikka, Vol. 30 Nro 1, 1997, s

ADAPTIVE APPROACH IN NONLINEAR CURVE DESIGN PROBLEM. Simo Virtanen Rakenteiden Mekaniikka, Vol. 30 Nro 1, 1997, s ADAPTIVE APPROACH IN NONLINEAR CURVE DESIGN PROBLEM Simo Virtanen Rakenteiden Mekaniikka, Vol. 30 Nro 1, 1997, s. 14-24 ABSTRACT In recent years considerable interest has been shown in the development

More information

MINIMIZING CONTACT STRESSES IN AN ELASTIC RING BY RESPONSE SURFACE OPTIMIZATION

MINIMIZING CONTACT STRESSES IN AN ELASTIC RING BY RESPONSE SURFACE OPTIMIZATION MINIMIZING CONTACT STRESSES IN AN ELASTIC RING BY RESPONSE SURFACE OPTIMIZATION Asim Rashid THESIS WORK 21 PRODUCT DEVELOPMENT AND MATERIALS ENGINEERING Postal Address: Visiting Address: Telephone: Box

More information

Chapter Multidimensional Gradient Method

Chapter Multidimensional Gradient Method Chapter 09.04 Multidimensional Gradient Method After reading this chapter, you should be able to: 1. Understand how multi-dimensional gradient methods are different from direct search methods. Understand

More information

1 2 (3 + x 3) x 2 = 1 3 (3 + x 1 2x 3 ) 1. 3 ( 1 x 2) (3 + x(0) 3 ) = 1 2 (3 + 0) = 3. 2 (3 + x(0) 1 2x (0) ( ) = 1 ( 1 x(0) 2 ) = 1 3 ) = 1 3

1 2 (3 + x 3) x 2 = 1 3 (3 + x 1 2x 3 ) 1. 3 ( 1 x 2) (3 + x(0) 3 ) = 1 2 (3 + 0) = 3. 2 (3 + x(0) 1 2x (0) ( ) = 1 ( 1 x(0) 2 ) = 1 3 ) = 1 3 6 Iterative Solvers Lab Objective: Many real-world problems of the form Ax = b have tens of thousands of parameters Solving such systems with Gaussian elimination or matrix factorizations could require

More information

Camera calibration. Robotic vision. Ville Kyrki

Camera calibration. Robotic vision. Ville Kyrki Camera calibration Robotic vision 19.1.2017 Where are we? Images, imaging Image enhancement Feature extraction and matching Image-based tracking Camera models and calibration Pose estimation Motion analysis

More information

Numerical Linear Algebra

Numerical Linear Algebra Numerical Linear Algebra Probably the simplest kind of problem. Occurs in many contexts, often as part of larger problem. Symbolic manipulation packages can do linear algebra "analytically" (e.g. Mathematica,

More information

THE DEVELOPMENT OF THE POTENTIAL AND ACADMIC PROGRAMMES OF WROCLAW UNIVERISTY OF TECH- NOLOGY ITERATIVE LINEAR SOLVERS

THE DEVELOPMENT OF THE POTENTIAL AND ACADMIC PROGRAMMES OF WROCLAW UNIVERISTY OF TECH- NOLOGY ITERATIVE LINEAR SOLVERS ITERATIVE LIEAR SOLVERS. Objectives The goals of the laboratory workshop are as follows: to learn basic properties of iterative methods for solving linear least squares problems, to study the properties

More information

The exam is closed book, closed notes except your one-page (two-sided) cheat sheet.

The exam is closed book, closed notes except your one-page (two-sided) cheat sheet. CS 189 Spring 2015 Introduction to Machine Learning Final You have 2 hours 50 minutes for the exam. The exam is closed book, closed notes except your one-page (two-sided) cheat sheet. No calculators or

More information

5 Machine Learning Abstractions and Numerical Optimization

5 Machine Learning Abstractions and Numerical Optimization Machine Learning Abstractions and Numerical Optimization 25 5 Machine Learning Abstractions and Numerical Optimization ML ABSTRACTIONS [some meta comments on machine learning] [When you write a large computer

More information

Alternating Projections

Alternating Projections Alternating Projections Stephen Boyd and Jon Dattorro EE392o, Stanford University Autumn, 2003 1 Alternating projection algorithm Alternating projections is a very simple algorithm for computing a point

More information

Optimization. Industrial AI Lab.

Optimization. Industrial AI Lab. Optimization Industrial AI Lab. Optimization An important tool in 1) Engineering problem solving and 2) Decision science People optimize Nature optimizes 2 Optimization People optimize (source: http://nautil.us/blog/to-save-drowning-people-ask-yourself-what-would-light-do)

More information

1.00 Lecture 19. Numerical Methods: Root Finding

1.00 Lecture 19. Numerical Methods: Root Finding 1.00 Lecture 19 Numerical Methods: Root Finding short int Remember Java Data Types Type byte long float double char boolean Size (bits) 8 16 32 64 32 64 16 1-128 to 127-32,768 to 32,767-2,147,483,648 to

More information

1. Introduction. performance of numerical methods. complexity bounds. structural convex optimization. course goals and topics

1. Introduction. performance of numerical methods. complexity bounds. structural convex optimization. course goals and topics 1. Introduction EE 546, Univ of Washington, Spring 2016 performance of numerical methods complexity bounds structural convex optimization course goals and topics 1 1 Some course info Welcome to EE 546!

More information

Full waveform inversion by deconvolution gradient method

Full waveform inversion by deconvolution gradient method Full waveform inversion by deconvolution gradient method Fuchun Gao*, Paul Williamson, Henri Houllevigue, Total), 2012 Lei Fu Rice University November 14, 2012 Outline Introduction Method Implementation

More information

Stopping Criteria for Iterative Solution to Linear Systems of Equations

Stopping Criteria for Iterative Solution to Linear Systems of Equations Stopping Criteria for Iterative Solution to Linear Systems of Equations Gerald Recktenwald Portland State University Mechanical Engineering Department gerry@me.pdx.edu Iterative Methods: High-level view

More information

Understanding Gridfit

Understanding Gridfit Understanding Gridfit John R. D Errico Email: woodchips@rochester.rr.com December 28, 2006 1 Introduction GRIDFIT is a surface modeling tool, fitting a surface of the form z(x, y) to scattered (or regular)

More information

Parallelization Strategy

Parallelization Strategy COSC 335 Software Design Parallel Design Patterns (II) Spring 2008 Parallelization Strategy Finding Concurrency Structure the problem to expose exploitable concurrency Algorithm Structure Supporting Structure

More information

Optimal Control Techniques for Dynamic Walking

Optimal Control Techniques for Dynamic Walking Optimal Control Techniques for Dynamic Walking Optimization in Robotics & Biomechanics IWR, University of Heidelberg Presentation partly based on slides by Sebastian Sager, Moritz Diehl and Peter Riede

More information

Adaptive Filtering using Steepest Descent and LMS Algorithm

Adaptive Filtering using Steepest Descent and LMS Algorithm IJSTE - International Journal of Science Technology & Engineering Volume 2 Issue 4 October 2015 ISSN (online): 2349-784X Adaptive Filtering using Steepest Descent and LMS Algorithm Akash Sawant Mukesh

More information

Modified Augmented Lagrangian Coordination and Alternating Direction Method of Multipliers with

Modified Augmented Lagrangian Coordination and Alternating Direction Method of Multipliers with Modified Augmented Lagrangian Coordination and Alternating Direction Method of Multipliers with Parallelization in Non-hierarchical Analytical Target Cascading Yongsu Jung Department of Mechanical Engineering,

More information

Truncation Errors. Applied Numerical Methods with MATLAB for Engineers and Scientists, 2nd ed., Steven C. Chapra, McGraw Hill, 2008, Ch. 4.

Truncation Errors. Applied Numerical Methods with MATLAB for Engineers and Scientists, 2nd ed., Steven C. Chapra, McGraw Hill, 2008, Ch. 4. Chapter 4: Roundoff and Truncation Errors Applied Numerical Methods with MATLAB for Engineers and Scientists, 2nd ed., Steven C. Chapra, McGraw Hill, 2008, Ch. 4. 1 Outline Errors Accuracy and Precision

More information

An Iterative Convex Optimization Procedure for Structural System Identification

An Iterative Convex Optimization Procedure for Structural System Identification An Iterative Convex Optimization Procedure for Structural System Identification Dapeng Zhu, Xinjun Dong, Yang Wang 3 School of Civil and Environmental Engineering, Georgia Institute of Technology, 79 Atlantic

More information

MAE 384 Numerical Methods for Engineers

MAE 384 Numerical Methods for Engineers MAE 384 Numerical Methods for Engineers Instructor: Huei-Ping Huang office: ERC 359, email: hp.huang@asu.edu (Huei rhymes with way ) Tu/Th 9:00-10:15 PM WGHL 101 Textbook: Numerical Methods for Engineers

More information

Ellipsoid Algorithm :Algorithms in the Real World. Ellipsoid Algorithm. Reduction from general case

Ellipsoid Algorithm :Algorithms in the Real World. Ellipsoid Algorithm. Reduction from general case Ellipsoid Algorithm 15-853:Algorithms in the Real World Linear and Integer Programming II Ellipsoid algorithm Interior point methods First polynomial-time algorithm for linear programming (Khachian 79)

More information

Logistic Regression

Logistic Regression Logistic Regression ddebarr@uw.edu 2016-05-26 Agenda Model Specification Model Fitting Bayesian Logistic Regression Online Learning and Stochastic Optimization Generative versus Discriminative Classifiers

More information

Machine Learning / Jan 27, 2010

Machine Learning / Jan 27, 2010 Revisiting Logistic Regression & Naïve Bayes Aarti Singh Machine Learning 10-701/15-781 Jan 27, 2010 Generative and Discriminative Classifiers Training classifiers involves learning a mapping f: X -> Y,

More information

Journal of Engineering Research and Studies E-ISSN

Journal of Engineering Research and Studies E-ISSN Journal of Engineering Research and Studies E-ISS 0976-79 Research Article SPECTRAL SOLUTIO OF STEADY STATE CODUCTIO I ARBITRARY QUADRILATERAL DOMAIS Alavani Chitra R 1*, Joshi Pallavi A 1, S Pavitran

More information

Optimization Plugin for RapidMiner. Venkatesh Umaashankar Sangkyun Lee. Technical Report 04/2012. technische universität dortmund

Optimization Plugin for RapidMiner. Venkatesh Umaashankar Sangkyun Lee. Technical Report 04/2012. technische universität dortmund Optimization Plugin for RapidMiner Technical Report Venkatesh Umaashankar Sangkyun Lee 04/2012 technische universität dortmund Part of the work on this technical report has been supported by Deutsche Forschungsgemeinschaft

More information

Ch 09 Multidimensional arrays & Linear Systems. Andrea Mignone Physics Department, University of Torino AA

Ch 09 Multidimensional arrays & Linear Systems. Andrea Mignone Physics Department, University of Torino AA Ch 09 Multidimensional arrays & Linear Systems Andrea Mignone Physics Department, University of Torino AA 2017-2018 Multidimensional Arrays A multidimensional array is an array containing one or more arrays.

More information

Linear Discriminant Functions: Gradient Descent and Perceptron Convergence

Linear Discriminant Functions: Gradient Descent and Perceptron Convergence Linear Discriminant Functions: Gradient Descent and Perceptron Convergence The Two-Category Linearly Separable Case (5.4) Minimizing the Perceptron Criterion Function (5.5) Role of Linear Discriminant

More information

Convexization in Markov Chain Monte Carlo

Convexization in Markov Chain Monte Carlo in Markov Chain Monte Carlo 1 IBM T. J. Watson Yorktown Heights, NY 2 Department of Aerospace Engineering Technion, Israel August 23, 2011 Problem Statement MCMC processes in general are governed by non

More information

Optimization Algorithms, Implementations and. Discussions (technical report for self-reference)

Optimization Algorithms, Implementations and. Discussions (technical report for self-reference) Optimization Algorithms, Implementations and Discussions (technical report for self-reference) By : Lam Ngok Introduction In this preliminary optimization study we tested and implemented six different

More information

Data Mining Chapter 8: Search and Optimization Methods Fall 2011 Ming Li Department of Computer Science and Technology Nanjing University

Data Mining Chapter 8: Search and Optimization Methods Fall 2011 Ming Li Department of Computer Science and Technology Nanjing University Data Mining Chapter 8: Search and Optimization Methods Fall 2011 Ming Li Department of Computer Science and Technology Nanjing University Search & Optimization Search and Optimization method deals with

More information